Skip to content
Bosch – Hemoglobin threshold-based RBC Transfusion during critical illness episode artwork

Oct 11, 2023 42 min

Bosch – Hemoglobin threshold-based RBC Transfusion during critical illness

from Maryland CC Project

Nick Bosch, MD, MSc is an assistant professor in the pulmonary division at the Boston University School of Medicine. In his lecture, he reviews the results of prior studies of […]

Episode details come from the podcast's RSS feed.

Embed this episode

Ready to play

Bosch – Hemoglobin threshold-based RBC Transfusion during critical illness

0:00 42:10
of matches

Transcript Auto-generated, may contain errors

It's my absolute pleasure to welcome back for Critical Care Grand Rounds, Nick Bosch. So as you just heard us talking about, we had chatted in the past about the use of phenobarbital, and I think he was pretty instrumental in us at the University of Maryland reassessing our paradigm of how we treat people in alcohol withdrawal. I read a paper of his recently that made me start to think a lot about hemoglobin and hemoglobin threshold and this teaching that we've had in critical care for as long as I can remember that you transfuse people at a hemoglobin of seven without really thinking more about the physiology or the downstream implications of that. And so I was absolutely delighted to have Nick here today to kind of share his work. He's an assistant professor at Boston University in their pulmonary division at the Pulmonary Center, and it's my great pleasure to have you.

them here talking today about phenoglobin thresholds in our critically ill patients thank you so much for being here again today Thanks, Andy. I'm excited. I have a cold, so I'll do my best to not lose my voice. Here are my disclosures. And then this is what we'll talk about today. I'm going to go a bit into stats this time because I think it's a really, really cool method and a method that has potential applications and lots of critical illness-related questions. So I think it's a cool method to talk about. And we'll talk about how our findings fit into the context of what's been found before. So this isn't news to anybody, right? Almost all patients, it's close to 100% of patients who walk into the ICU, leave the ICU with anemia.

Huge public health concern with this. There's millions and millions of units of blood transfused for patients who are in the ICU. And we ourselves are somewhat causing this by taking blood draws from them once a day, twice a day, three times a day. So almost all patients in the ICU have anemia. And that's been known for a long time. And seminal work by the TRIC group, this is 1999, were interested in looking at whether patients who have anemia have bad outcomes. And so this U-shaped curve demonstrates the relationship between hemoglobin on the x-axis and adjusted risk of mortality on the y-axis among critically ill patients. And this U-shaped graph is the same as we see for many times we measure physiology-based measures, right? Too high is bad, too low is bad, and something in the middle is probably best. And so these investigators at the time said, well, we're currently transfusing patients at a hemoglobin to keep their hemoglobin around 10.

But if we look at what a hemoglobin of nine is and a hemoglobin of 11, the mortality rate is pretty flat for that, for those groups of patients with those hemoglobin levels in this observational study. So these investigators said, well, what about what if we choose somewhere that's more steeply on the curve here where giving blood could be more likely to affect mortality? And so that's exactly what these investigators did. This is the landmark trial that put this number of seven in everyone's brain. So this is the TRIC trial a couple years after that observational study I showed you, and they compared what a restrictive transfusion strategy, start to transfuse when your hemoglobin is less than 7 versus start to transfuse when your hemoglobin is less than 10, and then there were some additional parameters. I don't think widely known, but this study had multiple changes in sample size because they couldn't enroll the number of patients they enrolled, and they ended up stopping enrollment early. Be that as it may, the results strongly suggested that a restrictive transfusion threshold was not only not harmful, here's 30-day mortality of 19 versus 23%, but it had a signal towards benefit.

And several subsequent RCTs testing in huge eye bleed, septic shock, hip fracture, all pretty much show the same thing, that either getting less blood by a restrictive strategy is better or at least not different than a liberal strategy. And there's been many meta-analyses, too, looking at whether we should be transfusing at lower or higher hemoglobin thresholds. And the takeaway is that it probably doesn't matter for mortality. And you can see here that this is the bottom of a force plot from a Cochrane review. And because it probably doesn't matter for mortality, we think that...

giving less blood in general as a community is better than giving more blood for many reasons, not the least of which it's a precious resource. And so surviving sepsis, some other critical care guidelines all say roughly the same thing. They say you should only transfuse when your hemoglobin is less than seven grams per deciliter because it's not harmful to do that practice. And I think this recommendation is very careful to say you should only transfuse when your hemoglobin is less than seven, not that you should always transfuse. And that's the difference between what we do in clinical practice, at least what we did here and what I think we continue to do here is...

do we always transfuse less than seven or do we only transfuse when it's less than seven? And those are different. And I know in my practice, you know, you get to rounds in the morning and everybody who had a 4 a.m. lab has already received a transfusion if a hemoglobin is less than seven. So that led us to think that there's a knowledge gap here. And the knowledge gap is this, right? And this is the problem with doing comparative efficacy or comparative effectiveness studies before placebo-controlled studies is we think on the spectrum of good to bad, transfusing less is probably better than transfusing at a higher threshold. But no one really knows whether you should maybe not transfuse specifically at a hemoglobin of 7. Maybe there's a better strategy. So we wanted to answer the question, does transfusion versus no transfusion at a hemoglobin of 7 affect outcomes? And so here, I think, put a different way.

If your low blood counts are bad, is raising blood counts when your hemoglobin is low good? And that's not clear, right? All treatments we give have risks. And blood, specifically someone else's blood who donated blood maybe weeks or months ago, doesn't mean it's good for the patients. And there are many risks that we know about. And it turns out that the benefits of blood transfusion, especially in critical illness, are less clear because we think that that's improving oxygen delivery, helping with CO2 removal. But especially in patients with shock, we may not actually be getting those benefits. So here is the study question that we sought to answer.

This is a little cheeky. I don't know if I really think this is true, but, you know, doing a trial usually requires some equipoise. And I think before our study, at least, it wasn't clear there was equipoise about withholding transfusions. And I think the point is, if you're saying you're going to withhold the elixir of life from patients who are critically ill, you may have a little bit of a tough time doing a clinical trial. I don't think that's necessarily the case here, but anymore, I do think we need a trial. But typically, you know, we're talking about treatments where there's equipoise that we want to do an RCT.

And so if we were to do an RCT to answer the study question, is giving blood at a hemoglobin threshold of seven better than placebo, we might be looking at a causal pathway like this. So this is called a directed acyclic graph. And in the causal inference literature, these arrows are intentionally one way, and they represent what we think are the relationships between variables. And so an RCT is testing this, right? And there's lots of other things that can affect outcomes. Let's say it's death, right? So being severely volume overloaded, of course, affects your outcomes in the ICU. But the important part of an RCT is that there's no link. There's no...

Volume overload can never cause red blood cell transfusion because that's due to randomization, right? That's the inherent benefit of RCTs. So even though there are patients who are volume overloaded in an RCT, they're balanced between groups because it's due to chance. And that is one of the inherent problems of observational studies. And I think this is the beginning of talking about the methods that we used to get around this is that volume overload may have an important impact on whether someone does receive a red blood cell transfusion, right? If you have a roaring florid heart failure exacerbation, people may be more cautious about giving you blood.

And it turns out that the measurable confounders, these are called confounders, are usually not that big of an issue. It's the fact that there's almost always other things that we either can't measure or don't even know to measure that also affect this. That's the inherent problem of all observational studies. And so, like I said, when you can measure confounding, there's tons of ways. For decades, we've had ways to account for that. There's adjustment, stratification, matching, propensity score. Now people are using targeted maximum likelihood estimation, all these machine learning, but they all have the same inherent problem is that there could be the risk of unmeasured confounding.

as shown here. So that's where my passion comes in. I think these are quasi-experimental study designs, and these are designs that also have many assumptions about them. You have to have a specific set of circumstances, a specific scenario has to exist. But within that circumstance, done well, you can actually minimize unmeasured confounding. And what all these studies have in common is that they, in some way, result in pseudo-randomization. And the strongest of those is one, and in my opinion, I'm going to talk about today is regression discontinuity, that in many cases can result in very close to what we think of as randomization from the observation.

And these are fun to think about as clinicians and researchers because they all require you to know something about what happens in real life. So understanding that there is, for example, for regression discontinuity, a specific number that we initiate treatment on, like hemopharm. So here's RD in a nutshell. So I'll let you read the official definition. The nice thing about RD is it's very graphical and easy to visualize. And so for RD to work, there has to be a clinical scenario that has this relationship or something similar. There has to be a running variable, in this case, the hemoglobin level, at which there's a sharp change in the likelihood of getting a treatment, right? And in this scenario...

especially if there is error in measuring hemoglobin. We all know that's the case, right? You get two hemoglobins in the same patient five minutes apart and they're gonna be slightly different. When that's the case, a patient who has a hemoglobin of 7.1 is likely very, very, very similar to a patient that has 6.9. Even if you repeated the test, they may have flip-flopped. And yet because of that hemoglobin level, the patients are treated versus not treated. And so that is the inherent benefit of regression discontinuity is that for patients very close to a threshold, they're basically randomized.

We all know in clinical practice, this isn't how the real world works. Some people with hemoglobin higher get transfused and some don't. And this just has a slightly different interpretation for regression discontinuity. The world's not perfect, but it works out nicely. It turns out that this fuzzy design, it's actually called fuzzy, is very close to an intention to treat trial. So we wanted to treat, but some patients don't end up getting their assigned treatment. So in regression discontinuity, analyzing these situations that aren't perfect end up with an intention to treat effect.

Once you know that there's the clinical situation where treatment has a large difference in probability across the threshold, you can then measure outcomes across that threshold. So if this situation exists, then you can look and see if there's an abrupt change in outcome. Maybe this is death, right? And so with these three lines, the middle line suggests that this treatment does not change outcome. The upper line suggests that it worsens outcomes, and the bottom line suggests it improves outcomes.

So that's the fun graphical part of regression discontinuity that you can really just see by looking at graphs, whether or not a treatment has discontinuity at a threshold and then whether outcomes change. So there are limitations. It turns out if you can do an RCT, you should do an RCT because you have to have a bunch of patients with hemoglobins that are really, really close to the cutoff value. So you have less power. And importantly, it only gives you evidence around the threshold itself. This method cannot be used to say anything about whether it would be better to transfuse patients at a hemoglobin less than six, for example. And that's, I mean, that's like the next step, right? Once you hear these results, I think the question is, okay, so what? So when do we transfuse patients?

And then lastly, you are probably hearing this word more and more, and this is what we did for our study, but target trial emulation is what observational health services researchers are adopting more and more as ways to minimize some of the mistakes we frequently make with observational studies. And so the idea behind this isn't too fancy. It's just designing your study as if it were in RCT. So with that idea, what target trial emulation does is you just specify the same sort of things that you would do in RCT, like who's eligible. For example, are you looking at ICU patients within a certain time period, a certain time from admission? What are the treatment strategies? What are the assignment procedures, et cetera? The target trial emulation is independent of the statistical analysis plan. So you can do a target trial emulation with any sorts of matching propensity score, quasi-experimental designs. The point is to set up the design of the study, not the stats of the study, to minimize biases.

And so this is a table that compares what we did to what the target trial that we wanted to do. For both the target trial and the observational study, we wanted to include people who were admitted to the ICU. And we were interested in comparing blood transfusion or placebo in a target trial. There's no placebo in the real world. So the comparison here is with usual care or no blood transfusion. In a target trial, it would be randomized. We didn't have that ability, so this is per-treated. And here's where the differences emerge a little bit, and I want to make this clear. I think in an ideal standpoint, you want to follow patients until they're discharged or 30 days, and you want to look at something like death.

In the RD framework, there's no way to specify that we want a treatment regimen that every time someone's hemoglobin drops less than seven, we give them transfusion versus placebo. RD is limited to a single transfusion. Or... the equivalent of repeating the randomization process each time someone has a hemoglobin less than seven. So because of that, we limited our follow-up period to a shorter time period because we were really interested in the effects of a single transfusion on outcomes rather than a policy of transfusing every time hemoglobin was low. And similarly, because we didn't really think that one transfusion would be strongly associated with death, our primary outcome was a modified SOFA score, so a measure of organ dysfunction that also accounted for the risk of death, and I'll show that in a second.

Moving on. So this was the study we published. These are the methods. We took three large clinical databases. MIMIC is a single center granular database that has 40,000 ICU patients. EICU is 208 hospitals across the U.S. that use telehealth, that use Phillips telehealth system. It's getting a little old now. It's from 2015. And then Premier is 25% of all hospitalized patients in the U.S., so a giant database. And we use all these databases separately to see if we got the same estimates in all three of these studies. We included adults.

We limited those patients with hemoglobin between days two and 28. We weren't interested in hemoglobin values on day one because people get resuscitated. They may have acute GI bleeds, reasons to get blood transfusion that isn't entirely based on hemoglobin. And then we, in our primary analysis, only looked at one hemoglobin per patient. In subsequent analyses, we considered repeated measures, that design where each time you were randomized, each time you had a hemoglobin measure.

The outcomes we looked at, we wanted to first know whether or not you got a blood transfusion based on your hemoglobin threshold. We wanted to know what patient's next hemoglobin was. So did blood transfusion do its intended effect? Did it raise your blood count? And then this was the primary outcome, this change in modified SOFA score. And so for those of you who don't know, your SOFA score is the same thing that's used in the sepsis 3 definition, and it's made up of parameters specific to different organs. For respiratory, it's PDAF ratio.

in use of mechanical ventilation, MAP and phasopressors for cardiac, et cetera. And what we did is we, the total score before death is 24. And we said, well, if you die during that time period, you get a score of 25. And that's similar to methods used to define things like ICU-free days, ventilator-free days that account for the competing risk of death, right? We don't want people who die to seem like they're less ill. So this is why we used this score.

Can I ask you a question about the SOFA score before you go further? Yeah. How do you, because I struggle with this in my research, a lot of the patients don't have bilirubin levels, they don't have ASTs, or they don't have a PAO2, so we're imputing S to F ratio really here. And I think the GSS, like your CNS evaluation is really not particularly reliable, especially in a sedated patient, a paralyzed patient. So how did you sort of deal with these limitations in your primary outcome here?

Yeah, it's a great question. And SOFA score is notorious for the different methods that are imputed for it. So, for example, the original derivation of the SOFA score used median and mean values, and they also had access to data from prior to admission SOFA scores. So there is a huge, I think that no one knows the right way to impute data for these scenarios. We said that if you did not have a measurement, it was normal.

That's not necessarily right, but I don't think anyone knows what's right. And, you know, imputed data with observational data is a huge issue. The thinking on it has, I think, changed over time. I think now we mostly realize that assuming a lab test isn't ordered, then it being missing is something that the clinician didn't have access to and likely wouldn't be a confounder in most cases. It's a different story if the reason that's missing is because of database problems. But if it's missing because it wasn't measured, then those likely don't influence treatment decisions because the clinician couldn't have used it.

This is just like schematic for one patient. So patients, we've randomly selected a hemoglobin per patient. So that could be from day two to day 28. On the day they had their hemoglobin checked, we looked to see whether they got a red blood transfusion, yes or no. And then we looked at their change in their SOFA score in the three days after. It sounds simple, but it turns out to be a lot messier than you might think. So this is made up data, but this is sometimes how the data actually looks.

And one of the problems with regression discontinuity in the assumptions is that we know what the true relationship is between hemoglobin and outcome. And that's just not true. No one knows what that true relationship is. And so you run several sensitivity analyses that try to draw different lines between different points. And you're hoping to see that these findings are similar no matter how you draw your line. And lines are just models. In research, lines connecting points or trying to minimize variation between two variables are just models. And so in the primary...

approach, we use local linear regression. And that's a fancy way of saying that we drew really short lines in a subset of the data that was really close to the threshold. But there's lots of other ways you could have done this. And this table just outlines some of the other alternatives that we also considered and used in our study. In RD, this difference in the outcome at the threshold is the effect estimate itself. So this is your absolute risk difference or your mean difference in outcomes between tree-ridden groups. It's just this distance.

All right, and you'll see what it looks like in a second. So this is our study inclusion. As you can see, the Premier database was overwhelmingly larger than these other studies, and I'm highlighting the results from that because they're less noisy. This is a typical table one for an RD study. You have columns for each database, MIMIC, EICU, and Premier. So just look at the Premier on the corner here. These are characteristics of the patient's average, right? So an age of 66, 45% female. And then what is important to do for RD is to see if these same characteristics are different at the threshold, right? So if the percent of patients that are female changes strongly at the threshold, then you should be concerned.

that the assumptions around RD, that the only thing that happens at that threshold is a change in the use of blood transfusions, you're worried about that assumption holding. And so in this case, you can see that for all of these variables we looked at, that there was no difference, no significant difference in any of them at the threshold, suggesting that nothing else is happening at the same time. It's not like at a hemoglobin 7, all of a sudden people get vasopressors and the use of vasopressors could explain outcomes. It looks like nothing else is happening. This is the visual graph of what I just told you. So here's age, sex, sulfa score before the blood transfusion, a measure of comorbidities, use of mechanical ventilation, use of vasopressors. And you can see that these blue points are the averages at each hemoglobin level, 7.1, 7.2, 7.3. And then the red lines are the model results. And you can see that basically there's very little difference, very little discontinuity in any of these variables at the threshold. What about outcomes? This summarizes what we found.

So when at the threshold, there was a large change in the people getting blood transfusion, that's what we expected to find, right? That's what had to be the case to do regression discontinuity. And that blood transfusion was also associated with an increase in your hemoglobin the next time it was checked. That also makes sense. So not only are you transfused, but your hemoglobin goes up after transfusion. And then we found that there was statistically no difference in outcomes of modified SOFA score from blood transfusion. All of them, all of the outcomes suggested that blood transfusion was harmful across the board. And among patients with sepsis, there was a significant increase in organ dysfunction associated with blood transfusion.

which we don't know, but we think that's because the microcirculatory changes in sepsis, this is completely a hypothesis, it could be a lie, means that the expected benefits from transfusion don't happen. So here is the difference in the probability of receiving treatment across the threshold. You can see it's about a 25% absolute risk difference. So it's not like all patients are getting transfused. And you can see it's only about 75% at best get transfused at hemoglobin less than seven. There's potentially many reasons for that. Religious preferences, CMO, basically goals of care differences. It could be issues with our ability to ascertain blood transfusion accurately, but this is what we found.

And you can see also that it's clearly fuzzy, right? There's many people who get transfusion even though their hemoglobin is higher, although that number goes way down less than eight. And this could be people with acute coronary syndrome. And I'll show you those subgroups in a second. This is the overall effect. So you can see here that there is a small jump in organ dysfunction at a hemoglobin threshold of seven. And it's about one point. So your SOFA score goes up by about one point from being transfused across the threshold. This is just showing the same relationship in all the data sets.

The left-hand side is the transfusion discontinuity. The right-hand side is the outcome discontinuity. Busy table, but just to say that on the right-hand side of the screen for Premier, all of the outcomes suggested harm. All of these are above one associated with blood transfusion. But again, few of them were statistically significant. The same relationship held true no matter how we looked at drawing our lines. We drew our lines, I don't know, 10 different ways, and all of the results were the same.

And then in subgroups, we found that there was a pretty strong effect that looked like among patients with sepsis, your organ dysfunction went up by about three points, and that was significant. Most patients in the cohort had sepsis. And then for cardiac ICU, we did not find any evidence of interaction. So it doesn't seem like if you have a primary cardiac problem that there's more benefit of getting transfused hemoglobin. We then wanted to understand what part of the SOFA score was going up.

And what we found, there were two that were statistically significant. Liver, which is bilirubin. So it's probably not liver. It's probably hemolysis. And then death. So death only went up a tenth of an absolute percentage point. But remember, trying to think that death in a 72-hour period after blood transfusion is affected by blood transfusion is a hard needle to move. So it's a small but significant increase in death. So we think that the results of our study...

This is Trick, right? Trick says seven is probably better than 10. We think that not giving blood at seven is probably better or at least not worse than seven. And this goes back to the underlying problem. What do you do with this data, right? If you can't let someone's blood drop to zero, you have to give them blood at some point. And I think that is the main limitation in our studies. It says that what we're doing now is probably not helping people, but what the alternative is is less clear.

And so I think this has opened the door for us to think about other studies, including testing lower thresholds, not using hemoglobin at all. Hemoglobin might not be the right measure. And there's others that have looked into other measures as well, like arterial venous difference, oxygen saturation difference, but none of those have panned out. But it doesn't tell us what to do clinically, I think. How low do you let it go? So these are some of the next steps that we are thinking about.

We're especially, I think, interested in this last one. We have some prelim data that we went back and stratified these results by sex. So women, starting at about, not with female patients, this isn't gender, this is sex, at about age 20, start to have lower hemoglobin counts than males. And that difference persists until old age, usually about 80 or 90. They finally equalize again. It's related to sex hormones. And they have lower baseline blood counts than males. And they have lower thresholds for anemia, as defined by who?

And so to us, that doesn't really make any sense why we would use a common threshold for all patients. It just doesn't make sense. And so in our prelim data, we did stratify this. And it looks like that both female and male patients are getting blood and hemoglobin 7. It doesn't look like people are treating those patients differently, but the harm is completely in women and men have no effect. So we're a little concerned about that. So that's kind of our next step study that we're hoping to flush out. And then I'll just end it with saying that.

I think everyone should be doing regression discontinuity. If you're an ICU researcher and ICU clinician, there's tons of things that you could apply regression discontinuity to. pH and bicarb administration, troponin level, and whether or not heparin drips are started. People have already studied D-dimer and risk of contrast-inducing nephropathy. It was a nice study that came out in JAMIM. We have this under review now, trying to see whether you give potassium. It's another thing we routinely do. You give potassium to get him up to 4 to 3.6. It's still under review and unpublished, but you can see that there's two discontinuities here. There's hospitals that replete at 4, and there's a bunch that replete at 3.6.

five and all these other ones that are kind of like, well, the lower it is, the more likely I am to give potassium. And then we took those hospitals and looked to see at either a 3.5 or four, whether that has any change. We use amiodarone use, which we are hoping is a marker for arrhythmia risk and found that there was no difference. So I think probably going to be more evidence that whether you transfer, whether you give potassium repletion at 3.5 versus four, it doesn't matter because those are both normal and unlikely to affect outcomes. And I think with that, it looks like there's a question, but I'll end. I want it to be short. It's a Friday. I think everyone probably doesn't want to have a long conversation. So here are some people to thank, especially Alan, who had been my mentor for a long time. And he's leaving. This is his last day at Boston University. He's moving to UMass, so we're all bummed. And that's it. And I'll try to answer this first question. So maybe I missed it. Yeah, so the question is, why can't the change in, why can't a hemoglobin less than seven be responsible? It's a good question. You do need to make the link that, you do need to.

You don't know. But with a large change in practice, the assumption is that's true. And the regression discontinuity analyses are really only saying something right at the threshold. So of course, lower hemoglobins are associated with bad outcomes and higher hemoglobins to a point are associated with better outcomes. But we're not interested in those extremes, those really low and really high. We're really interested in just the threshold. So unless you think that patients who have a hemoglobin of 7 versus 6.9 are different, are enough different, and that that could explain one to three point change in your SOFA score, then I think it would be hard to argue that.

Because again, we're talking about just at the threshold. And that's it. So any questions or comments? I think this presentation was great. I obviously invited you here for a reason. When I read your paper, I sort of thought to myself, like, this is going to...

change or challenge at least the paradigm by which we practice medicine and critical care in many ways, because I think this isn't an RCT and you're not claiming it's an RCT. There's lots of other studies, emulated trial designs and stuff in ECMO, kind of looking at ECMO patients versus non-ECMO patients, because we maybe don't feel good about doing RCTs, like you said, because there isn't the equipoise prior to the study. But I think to your point, like we've questioned, is seven the right number? We think about with title volume, six versus 12, but what's the right number in between? And so I think being able to at least get a better signal to kind of understand some of these seemingly arbitrary thresholds that we've set in medicine, just because that's how we design the RCTs to have two separate groups.

is a really cool and interesting idea that can then sort of say, okay, maybe we do need to study this in an RCT, right? Maybe we do need to assign, you know, seemingly, again, arbitrary groups, but one is seven and one is less than seven. And so I just think that this study design was particularly cool. And I think your reminder that blood transfusions are not benign, they can cause harm to patients, and that we're not seeing necessarily the benefit is a good one. So thanks for sharing that. I don't know if there were other questions from the audience here for Dr. Bosch.

Yes, this is Giorno. Great. Appreciate it. Really cool work. In terms of designing it, the patient population, I'm curious about why you included patients with hemorrhage in the analysis and whether that changed any of your point estimates.

Yeah, we considered acute GI bleed. There is evidence from one, I don't remember if it's the TRIC investigators, but it's the same study design, restrictive versus liberal transfusion in acute GI bleed. And we thought that in that study, it also shows a signal towards improved outcomes with a restrictive. We didn't think that we should exclude those patients specifically. We were interested in getting rid of their initial resuscitation phase.

by excluding day one because we thought that that was likely independent of hemoglobin level. And also when you, when we looked, we looked to see if there was a large change in transfusion on day one and there wasn't. People are basically just not using hemoglobin levels to decide whether to transfuse on that first day. We did not do a subgroup based on GI bleed and we probably should. It's a great point. So not a fan of subgroup analysis. I wasn't angling it that way, but definitely appreciate it. I'm curious also if we could just talk a little bit about, you know, one of the, one of the kind of conundrums in a lot of these transfusion studies is actually a moral time bias that.

The longer you're alive in the ICU, the more likely you are to get anemic and get transfused. And as you consider sort of these, you know, observational trial designs, what are ways to think or work around it? Yeah, I guess I would say for R&D specifically, I think at the threshold, there's relatively low risk for a moral time. We did run into, I think, heterogeneity of treatment effect or worries, but concern for heterogeneity of treatment effect with time. And when we initially submitted this, our primary analysis was a repeated measure. So every day, check their hemoglobin, see what the outcomes were. And there was concern that the effects of blood transfusion aren't the same early in someone's hospital course to later. That's not a moral time, but we did think about that. And that's why we ended up using a randomly selected per patient group, which should.

minimize that. Immortal time in observational research is a huge problem. And outside of RD, any comparison in observational research that's the usual care or no care should be, you should be very skeptical of because of the risk of immortal time and selection bias. So target trial emulation helps to reduce those. But anytime you, most observational studies that are done well from a causal effort standpoint should have an active control, not no treatment because of the risk of immortal time.

A mortal time is basically patients who live for a certain amount of time are more likely potentially to get treated because they had more time to accrue that time to be treated. And that's different from another group who doesn't have that time. And that leads to potential biased outcomes and is known to be a big problem in critical care. The combination of a mortal time and competing risk of death, again, patients in ICU die, are big barriers to overcome for all observational studies in ICU. You have another question in the chat. Did all the patients get PRVCs or did any patients get whole blood transfusions?

We lumped them together. Almost all were PRBCs, but we did not delineate. It's a good question. In the smaller cohorts, I don't think we would have been able to tell, and in the bigger cohort, yeah, we just did it, but it's a great point. Makes me think we should do this for platelets too, right? We should do it for platelet count of 50, 100, you know, whatever that threshold is that people decide that it's no longer safe.

I was thinking the same thing when you were presenting, and I feel like this is something maybe Giora taught me when he was my attending when I was in training, or I totally made this up, so sorry if I ascribe it to you, Giora. But I feel like that threshold is in this country, that in other countries, they perhaps don't use the same threshold of saying, we're going to transfuse everyone with platelets when their platelet count is 10. We're going to let them get to zero unless they're bleeding. I'm not sure why it's so different based on where you're caring for the patient, but you're right. I think trying to figure out what that right number should be is important.

Yeah, I think there's real power in study design to leverage that variation in practice with RD. And that's what we're doing with this potassium paper is different hospitals may do different cutoffs. And in traditional RD, you can only say things about one threshold. But if you have multiple thresholds, you can start to get an idea of the spectrum of effects depending on what threshold you choose. So that, I think, is potentially if there's variation in practice, but it's still defined by different thresholds, you can have some really powerful studies. And platelet seems like low-hanging fruit. We should do it. We could put it together. Are there other questions for Dr. Bosch?

All right. Well, I'm going to keep an eye on your platelet work then, since it sounds like that's the next thing that's coming your way. But I think similar to when you kind of challenged us here at Maryland to think about our paradigm about phenobarbital use, I think this may not completely change the way we practice medicine. It's not an RCT, but understanding its limitations, thinking about study design and thinking about all the other kind of arbitrary thresholds we set is really important. So I'm really happy you're here to share this with us today. And happy Friday. Have a good weekend. Thank you very much.

Beyond Hope Project with Jason Tharp Beyond Hope Project Welcome to 'Beyond Hope Project,' a beacon of inspiration in the podcast world. Join Jason Tharp as we embark on a transformative journey, exploring stories that awaken the will to start anew and find the strength to rise from setbacks. 'Beyond Hope Project' is about lighting the way to your starting line, not dictating your path. We delve into narratives that foster optimism and belief in a future where dreams become achievements. Each episode is a tapestry of positivity, enthusiasm, and comforting support, resonating with the essence of hope. This podcast goes beyond storytelling; it's a movement of positive change rooted in authenticity and transparency. We provide real tools for real journeys, infusing more hope and joy into the world. 'Beyond Hope Project' stands as a testament to the transformative power of hope and the limitless potential of the human spirit. Join us in transcending boundaries and breaking barriers. Here, we prove dai Veteran Chat Project Kyle Turner Our goal is to reach as many Veterans as we can. We noticed a lot of the podcast personalities tend to be unrelatable to the every day joe. Not taking anything away from anyone just hoping to give more people a place to feel connected to something. KZ and Kyle are Army vets who deployed together for OIF 07-09 and OND. Now close to 15 years later we are finding a way to open up about our issues, and find a way to build a community. Paramedic Philanthropy Project Melissa Ade Podcast by Melissa Ade Bigfoot Information Project Podcast Bigfoot Information Project Occasional conversations with leading researchers and investigators regarding the animal commonly referred to as bigfoot.

Frequently asked questions

How long is this episode of Maryland CC Project?

This episode is 42 minutes long.

When was this Maryland CC Project episode published?

This episode was published on October 11, 2023.

Is there a transcript available for this episode?

Yes, a full transcript is available for this episode. You can read the complete transcript on the episode page.

Can I download this Maryland CC Project episode?

Yes. Use the download control on the episode player to save the publisher-provided media file.